群随机研究中倾向性评分匹配与混合效应模型比较

赵航; 余小金

doi:10.16462/j.cnki.zhjbkz.2024.01.018

群随机研究中倾向性评分匹配与混合效应模型比较

doi: 10.16462/j.cnki.zhjbkz.2024.01.018

赵航,
余小金^,

东南大学公共卫生学院流行病与卫生统计学系，南京 210009

基金项目:

国家自然科学基金 81673274

详细信息

通讯作者:
余小金，E-mail：xiaojinyu@seu.edu.cn

中图分类号: R181.2
计量
- 文章访问数: 128
- HTML全文浏览量: 51
- PDF下载量: 18
- 被引次数: 0
出版历程
- 收稿日期: 2022-07-28
- 修回日期: 2022-09-30
- 网络出版日期: 2024-02-05
- 刊出日期: 2024-01-10

Comparison of propensity score matching with mixed-effect model in cluster randomized trial

ZHAO Hang,
YU Xiaojin^,

Department of Epidemiology and Health Statistics, School of Public Health, Southeast University, Nanjing 210009, China

Funds:

National Natural Science Foundation of China 81673274

More Information

Corresponding author: YU Xiaojin, E-mail: xiaojinyu@seu.edu.cn

摘要

摘要: 目的比较倾向性评分匹配(propensity score matching, PSM)法和混合效应模型法在群随机试验中的统计效能，为同类研究的统计分析方法选择提供指导。方法通过模拟研究与急性缺血性脑卒中疗效评级数据，比较倾向性评分后拟合单因素条件logistic回归分析模型和直接拟合混合效应模型应用于含有混杂因素的群随机试验数据时的统计学性能，说明方法应用场景和选择策略。结果 PSM后，各混杂因素组间均衡性明显改善。条件logistic回归分析模型和混合效应模型处理效应估计结果十分接近，但前者的P值更小。实例分析结果显示，匹配后各混杂因素的标准化均数差异(standardized mean difference, SMD)均控制在0.1以内，条件logistic回归分析模型识别出两个研究结局的处理组间差异；而混合效应模型仅识别出7 d有效率的处理组间差异。结论 PSM可以平衡群随机试验中的混杂因素，提高两组间的可比性，其检验效能较高。推荐在临床群随机试验中优先考虑PSM法，但也要注意其应用条件和局限性。
- 群随机试验 /
- 倾向性评分 /
- 混合效应模型
Abstract: Objective To compare the efficacy of propensity score matching (PSM) and mixed-effect model in statistical performance, providing methodological guidance for similar studies. Methods Employing a simulation study and a case study of acute ischemic stroke, this research compared the statistical performance of propensity score matching followed by univariate conditional logistic regression and mixed-effect model. It aims to illustrate their application scenarios and selection strategies in cluster randomized data involving confounders. Results In the simulation study, the balance of confounders between groups was significantly improved after PSM. Conditional logistic model provides similar treatment effect estimates but smaller P values compared with mixed-effect model. In the acute ischemic stroke case study, the standardized mean difference (SMD) of confounders were reduced below 0.1 after matching. Conditional logistic model identified the between-group difference in both outcomes. In contrast, mixed-effect model only identified the between-group difference in 7-day effective rate. Conclusions Propensity score matching can effectively improves groups comparability. Compared with mixed-effect model, conditional logistic model based on matched data has higher power of test. PSM is advisable for priority consideration in clinical cluster randomized trials, keeping in mind its specific application conditions and limitations.
- Cluster randomized trials /
- Propensity score /
- Mixed-effect model

HTML全文

图 1 匹配前后组间不均衡变量分布情况(mRS评分)

Figure 1. Distribution of unbalanced variables before and after matching (mRS score)

下载: 全尺寸图片幻灯片

表 1 模拟数据混杂变量分布

Table 1. Sampling distribution of confounding in the simulated data

混杂效应Mixed effect	分组Group	X1	X2	X3
低差异Low difference	对照组Control	N(0.0, 1.0)	N(0.0, 4.0)	Bernouli(0.45)
	试验组Treatment	N(0.3, 1.0)	N(0.6, 4.0)	Bernouli(0.30)
高差异High difference	对照组Control	N(0.0, 1.0)	N(0.0, 4.0)	Bernouli(0.60)
	试验组Treatment	N(0.6, 1.0)	N(1.2, 4.0)	Bernouli(0.30)

下载: 导出CSV

表 2 两种方法疗效指标分析

Table 2. Analysis of efficacy indicators of different methods

结局指标 Outcome	方法 Method	试验组发生人数(占比/%) Number of occurrences in treatment group(proportion/%)	对照组发生人数(占比/%) Number of occurrencesin control group(proportion/%)	OR值value (95% CI)	P值 value
NIHSS 7 d有效率 7-day NIHSS effective rate	混合效应模型 Mixed-effect model	372(73.96)	405(67.73)	1.404(1.074~1.836)	0.013
	PSM后条件logistic回归分析模型 Conditional logistic model	321(75.89)	282(66.67)	1.574(1.164~2.128)	0.003
90 d mRS评分≤2分 90-day mRS score≤2 score	混合效应模型 Mixed-effect model	381(78.56)	377(66.73)	1.459(0.939~2.265)	0.093
	PSM后条件logistic回归分析模型 Conditional logistic model	311(76.98)	285(70.54)	1.513(1.073~2.132)	0.018

下载: 导出CSV

参考文献(7)

[1]	黄丽红, 王永吉, 王素珍, 等. 倾向性评分方法及其规范化应用的统计学共识[J]. 中国卫生统计, 2020, 37(6): 952-958. DOI: 10.3969/j.issn.1002-3674.2020.06.041. Huang LH, Wang YJ, Wang SZ, et al. Statistical consensus on peopensity score methods and their normative application[J]. Chinese Journal of Health Statistics, 2020, 37(6): 952-958. DOI: 10.3969/j.issn.1002-3674.2020.06.041.
[2]	Lalani N, Jimenez RR, Yeap B. Understanding Propensity Score Analyses[J]. Int J Radiat Oncol, 2020, 107(3): 404-407. DOI: 10.1016/j.ijrobp.2020.02.638.
[3]	龚清海, 张晓宏, 徐琛玮. 混合效应模型在系统分组资料中的应用及SAS实现[J]. 中国卫生统计, 2009, 26(6): 577-579. DOI: 10.3969/j.issn.1002-3674.2009.06.005. Gong QH, Zhang XH, Xu CW. Mixed model in the hierarchical classification datas and implementation of SAS[J]. Chinese Journal of Health Statistics, 2009, 26(6): 577-579. DOI: 10.3969/j.issn.1002-3674.2009.06.005.
[4]	Austin PC. Optimal caliper widths for propensity-score matching when estimating differences in means and differences in proportions in observational studies[J]. Pharmaceutical Statistics, 2011, 10(2): 150-161 DOI: 10.1002/pst.433.
[5]	Zhang XT, Zhong WS, Ma XD, et al. Ginkgolide with intravenous alteplase thrombolysis in acute ischemic stroke improving neurological gunction: a multicenter, cluster-randomized trial (GIANT)[J]. Front Pharmacol, 2021, 12: 792136-792134. DOI: 10.3389/fphar.2021.792136.
[6]	Benedetto U, Head SJ, Angelini GD, et al. Statistical primer: propensity score matching and its alternatives[J]. Eur J Cardiothorac Surg, 2018, 53(6): 1112-1117. DOI: 10.1093/ejcts/ezy167.
[7]	Zhang ZN. Propensity score method: a non-parametric technique to reduce model dependence[J]. Ann Transl Med, 2017, 5(1): 7-14. DOI: 10.21037/atm.2016.08.57.